Accurate foot orientation and trajectory estimation are pivotal for advanced gait analysis, yet achieving this with inertial measurement units (IMUs) remains challenging due to their susceptibility to external acceleration, magnetic disturbances, and unbounded position errors. To address these limitations, we propose a multi-kernel correntropy smoother, which effectively mitigates unknown disturbances and enhances orientation accuracy. Furthermore, while the conventional zero-velocity update (ZUPT) method has been widely adopted in IMUs, the impact of incorporating position constraints has been largely overlooked. This paper demonstrates that by incorporating a single loop closure, the maximum positioning error can be reduced by up to 75%, with further reductions achievable through multiple position constraints. Comprehensive theoretical analysis and extensive experimental validation confirm the superior performance of the proposed methods.

Postprandial glucose management is crucial for patients with diabetes. However, the currently adopted meal-bolus decision algorithms normally rely on subject-specific parameters and are impacted by time-varying insulin requirements caused by behavioral or metabolic variations. In this work, a data-driven meal-bolus decision method with an insulin-sensitivity-aware event-triggered learning mechanism is proposed. Specifically, the algorithm is developed within the framework of adaptive dynamic programming, which utilizes a Gaussian process regression approach to construct the model network and applies two shallow neural networks to construct the critic and action networks. To allow the algorithm to quickly perceive and respond to the variation of glycemic dynamics, different utility functions characterized by insulin-sensitivity-dependent penalty terms are constructed and switched based on the estimated insulin sensitivity conditions. The effectiveness of the proposed method is evaluated by in silico experiments utilizing the 10-adult cohort from the FDA-accepted UVa/Padova T1DM simulator. For the scenario of decreased insulin sensitivity, the proposed method leads to statistically significant improvements in terms of the percent time in euglycemia 70-180 mg/dL (86.2% vs. 72.4%,$p=0.025$) and the mean glucose (142.3 mg/dL vs. 155.3 mg/dL,$p=0.020$) without increasing the risk of hypoglycemia compared with the standard bolus calculator. Besides, an advisor-mode analysis using clinical data for subjects who initially exhibited significant hyperglycemia upon starting insulin therapy is conducted, which proves its effectiveness and safety in practical applications. Note to Practitioners—This work is motivated by the issue of postprandial blood glucose management, which is indispensable to diabetes patients. An effective meal-bolus decision method can help individuals resist the surge in blood glucose levels caused by carbohydrate ingestions and reduce the risk of hyperglycemia. However, for the state-of-the-art bolus decision algorithms, the inter/intra-subject variations of physiological parameters and the fluctuating insulin requirements caused by physical activity, stress, or emotional change are the two main causes of performance degeneration. To this end, we propose an insulin-sensitivity-aware event-triggered learning approach to meal bolus optimization. Specifically, without explicitly adjusting the subject-specific parameters, an adaptive dynamic programming decision framework is designed to perform the bolus decision based on the data-driven network learning and the iteration of control law and cost function. Besides, the estimated insulin sensitivity information is utilized in the learning phase of the decision framework to optimize the ultimate bolus, bolus dosage so as to make the algorithm quickly respond to the variation of insulin requirements. The in silico experiments and clinical data advisor-mode analysis demonstrate the effectiveness of the proposed algorithm and its feasibility in clinical application for postprandial blood glucose management.

Recent advances in hybrid automated insulin delivery (AID) systems have demonstrated promising clinical potential for blood glucose regulation of patients with type 1 diabetes. However, the performance of the closed-loop glucose control algorithms in AID systems is still challenged by the substantial inter-subject variability. In this work, we propose a personalized controller based on an offset-free data-enabled predictive control (DeePC) method. Specifically, different Hankel matrices based on the subject-specific insulin-glucose trajectories for typical glycemic behavioral scenarios are constructed and scheduled based on the current glucose fluctuations to depict the highly nonlinear behaviors of the glucose dynamic system. An incremental form of insulin inputs is adopted in the formulation of DeePC to mitigate the tracking offsets caused by the mismatch in the basal rate profile or the behavioral system representation. Besides, based on the estimated individualized insulin sensitivity, an adaptive penalty mechanism is designed, which utilizes an insulin-sensitivity-dependent input penalty term in the optimization problem to adjust the aggressiveness of the control algorithm in real-time to adapt to the variation of the insulin requirements. The effectiveness of the proposed method is evaluated by the designed comprehensivein silicoexperiments from the FDA-accepted UVa/Padova T1DM simulator. For the scenario of under-estimated basal rate, the proposed algorithm obtains statistically significant improvement in terms of the percent time in euglycemia 70-180 mg/dL (73.4% vs. 64.8%) and the mean glucose (150.4 mg/dL vs. 166.4 mg/dL) in comparison with the model-based controller. Advisory-mode analysis using clinical data indicates that the proposed controller can mitigate persistent hyperglycemia by recommending additional insulin delivery dosages compared to those administered by the clinician.

OBJECTIVE: The limited labeled data hinders the application of medical artificial intelligence technology in the field of diabetes classification. In this paper, a pseudo-label supervised meta-learning algorithm supported by heterogeneous covariates data is proposed to implement diabetes classification tasks with fewer labeled samples. METHODS: First, clustering algorithms are employed to generate pseudo labels of samples, which are further used to create multiple pseudo-supervised tasks for meta-learning within the framework of few-shot learning. Second, the time and date features of dynamically monitored glucose data are extracted as dynamic covariates, while the physiological indicators from medical single sampling serve as static covariates. By incorporating these heterogeneous covariates, the model inputs are enriched from multiple perspectives, compensating for the homogeneity deficiency of data and providing complementary information. Finally, a pseudo-supervised meta-learning algorithm is proposed to learn the data features supported by heterogeneous covariates in a task-driven manner. The optimal model is then fine-tuned on downstream real diabetes classification tasks, enabling rapid adaptation to unseen new tasks. RESULTS: The proposed algorithm is thoroughly evaluated using clinical data, achieving an accuracy of 95.994% and an F1 score of 91.261%. CONCLUSION: The proposed method remains preferable for diabetes classification when compared to the state-of-the-art methods. SIGNIFICANCE: The approach offers an effective strategy for diabetes classification tasks with incomplete and limited labeled data.

In this study, the properties of inter-event times in probability for stochastic linear event-triggered control systems are explored. The analysis of inter-event intervals is conducted for three distinct classes of event-triggering mechanisms: the absolute, relative, and mixed ones. Given the inherent stochastic nature of systems, it is challenging to achieve a certain judgment on avoiding Zeno behavior, where events occur at an infinite frequency. To address this issue, a probabilistic approach to examine the inter-event times is employed. It enables us to quantify the likelihood of Zeno behavior occurring in different scenarios, providing valuable insights into the performance of stochastic event-triggered control systems. Our research reveals substantial differences in the likelihood on the occurrence of a positive minimum inter-event interval among different mechanisms. Specifically, the mixed event-triggering mechanism emerges as the most probable one to yield a positive minimum inter-event time, indicating its potential superiority in efficiency. Conversely, some solutions of relative event-triggered control are proved to exhibit Zeno behavior with a probability of 1. Finally, a numerical example is provided to illustrate the efficiency and feasibility of the obtained results.

This paper investigates the robustness and optimality of the multi-kernel correntropy (MKC) on linear regression. We first derive an upper error bound for a scalar regression problem in the presence of arbitrarily large outliers. Then, we find that the proposed MKC is related to a specific heavy-tail distribution, where its head shape is consistent with the Gaussian distribution while its tail shape is heavy-tailed and the extent of heavy-tail is controlled by the kernel bandwidth. Interestingly, when the bandwidth is infinite, the MKC-induced distribution becomes a Gaussian distribution, enabling the MKC to address both Gaussian and non-Gaussian problems by appropriately selecting correntropy parameters. To automatically tune these parameters, an expectation-maximization-like (EM) algorithm is developed to estimate the parameter vectors and the correntropy parameters in an alternating manner. The results show that our algorithm can achieve equivalent performance compared with the traditional linear regression under Gaussian noise, and significantly outperforms the conventional method under heavy-tailed noise. Both numerical simulations and experiments on a magnetometer calibration application verify the effectiveness of the proposed method.Note to Practitioners—The goal of this paper is to enhance the accuracy of conventional linear regression in handling outliers while maintaining its optimality under Gaussian situations. Our algorithm is formulated under the maximum likelihood estimation (MLE) framework, assuming the regression residuals follow a type of heavy-tailed noise distribution with an extreme case of Gaussian. The degree of the heavy tail is explored alternatingly using an Expectation-Maximization (EM) algorithm which converges very quickly. The robustness and optimality of the proposed approach are investigated and compared with the traditional approaches. Both theoretical analysis and experiments on magnetometer calibration demonstrate the superiority of the proposed method over the conventional methods. In the future, we will extend the proposed method to more general cases (such as nonlinear regression and classification) and derive new algorithms to accommodate more complex applications (such as with equality or inequality constraints or with prior knowledge of parameter vectors).

Time and fuel balance is an important topic in the dynamic tracking problem of unmanned systems. In this paper, we propose a quasi time-fuel optimal control strategy (QTFOC) to solve the dynamic tracking problem of a double integrator system, which is capable of multi-target switching tracking tasks, such as the multi-target strike of weapons and the rapid multi-target grabbing of robots on industrial assembly lines. Compared with the previous optimal control algorithms, the proposed controller retains the time-fuel optimal characteristic when switching among multiple dynamic targets, and overcomes the high-frequency oscillation problem by incorporating local linear control region and two nonlinear buffer areas. That is, when switching among multiple dynamic targets, the proposed control strategy enables the corresponding system to achieve desired transient performance and satisfactory steady-state performance simultaneously. In addition, the asymmetry of friction load is further explored, which affects the dynamic performance of the actual system. Extensive experiments based on the visual tracking turntable demonstrate the superiority and feasibility of the proposed method. Note to Practitioners—This paper was motivated by the problem of multi-target switching tracking, such as the multi-target strike of weapons with vision sensors and the rapid multi-target grabbing of robots on industrial assembly lines. In recent years, various algorithms have been developed for trajectory planning and tracking. However, it is still challenging for unmanned servo system to capture dynamic targets quickly due to the constrains of motion performance, energy, and computational power. In order to reduce the computational burden and obtain ideal response under various physical constraints, a quasi time-fuel optimal control strategy (QTFOC) with analytical solutions is proposed in this paper. First, the static target is extended to the dynamic target by further investigating the traditional time-fuel optimal control theory for double integrator system. Second, to deal with the oscillation problem caused by system disturbances, we incorporate buffer areas and local linear control region into the control algorithm, which makes the system more robust. Third, the system performance is further improved by analyzing the frictional load asymmetry. This algorithm requires small computational resources and can be implemented directly on microcontrollers such as STM32. The experimental results demonstrate the superiority of our proposed method in handling the multi-target switching tracking problem, which can also adjust the response speed weight, making the unmanned system perform better under different operating conditions. Furthermore, the proposed QTFOC can be extended to other unmanned systems, such as trajectory planning and motion control for unmanned vehicles and bionic robots.

In this work, an event-triggered learning problem for state-of-charge (SOC) estimation of long-term spacecraft li-ion batteries with system aging and disturbances is investigated based on a physics-informed long short-term memory (PI-LSTM) network. An equivalent circuit model and a pretrained Gaussian process regression model are integrated into a long short-term memory (LSTM) network, which is trained and updated quickly with limited transmission data. By considering noisy data and physical constraints simultaneously, the PI-LSTM approach provides interpretable dynamic models for the long-term battery SOC estimation. Then, an unscented Kalman filter is proposed to estimate the SOC performance. By using weighted average voltage prediction errors, an event-triggering condition is established to guarantee the estimation performance with a reduced signal transmission rate. The effectiveness of the proposed approach is validated through experiments on a real spacecraft Li-ion battery platform, which achieves the SOC estimation error of less than 2%, and the maximum voltage prediction error is reduced by 61% after updating the PI-LSTM model.

The rapid developments in Internet of Medical Things open up new avenues for personalized healthcare. Continuously monitored physiological data can be collected by wearable devices and are transmitted to a remote server for real-time monitoring and diagnosis. This article concerns a risk assessment problem of acute mountain sickness (AMS) with data transmitted according to an event-triggered transmission schedule. An event-triggered signal processing approach is introduced to reconstruct the untransmitted information, based on which, a dynamic SpO$_{\bf 2}$index (DSI) is further proposed for AMS risk evaluation. The performance of the proposed approach is analyzed through physiological data collected in a proof-of-the-concept study (N=12). Statistical significant correlation of the DSI with AMS ground truth including Lake Louise score, deep sleep duration, deep sleep ratio, and mean SpO$_{\bf 2}$during sleep is observed. More importantly, it is observed that the proposed event-triggered signal processing procedure can dramatically reduce the data transmission rate while maintaining the performance of the DSI assessment, through comparison of the DSI obtained using the proposed event-triggered approach with those obtained based on event-triggered raw data and continuously transmitted time-triggered data. The obtained results indicate the feasibility of adopting event-triggered data scheduling and signal processing to achieve AMS risk evaluation using data from wearable devices with limited communication/battery resources.

The design and implementation of an integrated wearable face recognition and training system for prosopagnosia patients are presented. The purpose of this assistive technology is to provide real-time memory assistance and long-term rehabilitation. The real-time face recognition mode provides audio and visual notification of people who interact with the subject, while the at-home training mode combines features of mnemonic and perceptual training to help with prosopagnosia rehabilitation. In addition, a custom eye tracker is developed to determine the person whom the subject is making eye contact with within a crowd. Using the inverted face effect to mimic the difficulties of prosopagnosia patients, clinically healthy participants have shown improvements in their face-naming abilities. Early results indicate the system's potential to enrich the well-being of prosopagnosia patients.